题　目：A Trustworthy Time-Watermarking Framework for AIGC Intellectual Property Dispute Resolution

嘉　宾：车尚锟，特聘副研究员，武汉大学科技商学院

主持人：张宏哲，助理教授，上海交通大学安泰经济与管理学院

时　间：2026年10月23日（周五）10:00-11:30

地　点：上海交通大学安泰经济与管理学院浩然高科技大厦306

内容简介:

The widespread use of Large Language Models (LLMs) in text generation has raised growing concerns about intellectual property disputes. Watermarking techniques, which aim to embed meta information into AIGC, have the potential to serve as judicial evidence. However， existing LLM watermarking methods rely on statistical signals embedded in token distributions, leading to inherently probabilistic detection. Their reliability further degrades when encoding multi-bit information (e.g. timestamps), often falling far below 100% accuracy. Moreover， these methods introduce detectable statistical patterns, making them vulnerable to forgery attacks, and also allow model providers to fabricate arbitrary watermarks. To address these limitations, we introduce the concept of trustworthy watermark, which enables reliable watermark recovery with 100% identification accuracy while resisting both user-side statistical attacks and provider-side forgery. We integrate several cryptographic techniques and develop a novel trustworthy time watermarking framework. Unlike existing approaches that directly encode time as the watermark payload, we embed time information into time-dependent secret keys under regulatory supervision, preventing provider-side forgery of arbitrary timestamps. The watermark payload is decoupled from time and instantiated as a randomly generated, non-stored bit sequence for each generation, eliminating statistical patterns and rendering statistical attacks ineffective. To address the unverifiability introduced by random payloads, we design a two-stage encoding mechanism that makes the recovered payload verifiable. Combined with error-correcting coding, this enables reliable recovery of generation time with theoretically perfect identification accuracy.

演讲人简介：

车尚锟现任武汉大学科技与产业学院、科技商学院特聘副研究员，他于2025年取得清华大学博士学位。车尚锟的研究聚焦于可信人工智能领域，使用计算设计科学研究方法，重点关注AI及大语言模型等技术的公平性（偏差）、可解释性、知识产权争议等社会问题，关注人力资源、金融、医疗等场景。研究成果发表于ACM TOIS、Scientific Reports、ICIS等国际期刊和会议。

欢迎广大师生参加！